Speech Recognition

Question 1

what are acoustic phonetics?

Answer 1

study of the physical properties of speech

Question 2

what is sound?

Answer 2

a vibration that propagates as an acoustic wave
(based on the perception of its characteristics)

Question 3

what is frequency?

Answer 3

the number of times per second a sound wave cycles from the highest to the lowest point.

Question 4

what is amplitude?

Answer 4

height of the wave
taller the wave = louder the sound

Question 5

what is a sound spectrogram?

Answer 5

is a visual representation of the spectrum of frequencies of sound

Question 6

axis’s of sound spectrogram?

Answer 6

frequency of sound on vertical axis, time on horizontal axis, intensity shown by darkness

Question 7

sound spectrogram formats?

Answer 7

dark bands (i.e., most intensity)
▪ Steady state formant (stays same over time)
▪ Formant transitions (changes over time)

Question 8

problems posed for speech recognition?

Answer 8

• lack of invariance
• problem in speaker variability
• segmentation problem

Question 9

what is lack of invariance?

Answer 9

no one-to-one correspondence between speech cues and perception

Question 10

what is the problem in speaker variability?

Answer 10

People differ in production of speech sounds –across people and occasions

Question 11

what is a segmentation problem?

Answer 11

people typically do not leave breaks between words when speaking.

Question 12

what is categorical perception?

Answer 12

We do not discriminate sounds within a phonemic category
- ex: we classify speech sounds as one phoneme or another

Question 13

modularity (revisited) and categorical perception?

Answer 13

• Some people have taken categorical perception as evidence for a speech perception module
• chinchillas show categorical perception

Question 14

what are some speech segment strategies?

Answer 14

• possible word constraint: tendency to segment speech so that each segment is a possible word
• Bilingual speakers tend to use strategies that are consistent with their dominant language

Question 15

what does context and speech recognition involve? when are people better at identifying words?

Answer 15

• people are better at identifying words when presented in sentences than when presented in isolation
• speech recognition involves bottom-up and top-down processing

Question 16

what are the two views of context and speech recognition?

Answer 16

• Autonomous view: context has effect after lexical access
• Interactionist view: allows context to affect earlier (lower) levels of processing

Question 17

context and speech recognition examples?

Answer 17

• When shadowing, Ps have a tendency to correct speech errors (e.g., Marslen-Wilson & Welsh)
• Phonemic restoration effect (e.g., Warren)
• Semantic & syntactic factors in speech perception (e.g., Miller & Isard)

Question 18

what does it mean when shadowing, Ps have a tendency to correct speech errors?

Answer 18

Ps are more likely to correct when the..
- context is highly predictable (role of semantic & syntactic factors)
- presented phoneme differed from the target phoneme by fewer distinctive features (role of bottom-up processing = gar v. car)

Question 19

what is the phonetic restoration effect?

Answer 19

the illusion that a phoneme deleted from a string of speech is actually there.
(ex: coughing replacing letter)

Question 20

what are semantic & syntactic factors in speech perception?

Answer 20

Ps shadow word strings in varying degrees of background noise
- 3 types of word strings:
* grammatical
* anomalous
* ungrammatical

Question 21

what is prosody?

Answer 21

tune and rhythm of speech

Question 22

Prosodic factors in speech recognition?

Answer 22

• Stress
• speech rate
• characteristics of individual speakers

Question 23

what is the mcgurk effect?

Answer 23

• Hear /ba/
• See /ga/
• Both –perceive /da/
  *** Importance of visual & auditory information for speech perception

Question 24

what are the two models of speech recognition?

Answer 24

• cohort
• TRACE

Question 25

what is cohort?

Answer 25

Spoken word recognition occurs in stages: 1. Access stage: set up initial cohort, strictly bottom-up process 2. Selection stage: words are eliminated from cohort until 1 item is left 3. Integration stage: the selected item is integrated into the representation of the utterance

Question 26

what is the selection stage based on?

Answer 26

- additional phonemic information - context of spoken sentence (early version of model)

Question 27

what part of the word do we pay more attention to?

Answer 27

the beginning - supported through selection stage

Question 28

what is the TRACE model?

Answer 28

take all of the various sources of information found in speech and integrate them to identify single words.

Question 29

what are connectionist models composed of?

Answer 29

- they contain a system of interconnected nodes - they have excitatory (facilitatory) and inhibitory connections between nodes - processing is massively parallel - has both top-down and bottom-up processing

Question 30

what is lexical access?

Answer 30

the retrieval of words from the mental lexicon, both in recognition and in production.

Question 31

what is the TRACE model composed of?

Answer 31

- connectionist model - 3 levels in the network: word, phoneme, and feature - incorporates top-down effects on the activation of features

Question 32

what is lexicon?

Answer 32

the vocabulary of a person, language, or branch of knowledge.

Question 33

examples of lexicon?

Answer 33

"No-hitter," "go-ahead run," and "Baltimore chop" are part of the baseball lexicon.

Question 34

example of prosidy?

Answer 34

"Yeah, that was a great movie," can mean that the speaker liked the movie or the exact opposite, depending on the speaker's intonation.

Question 35

what is coarticulation (lack of invariance)?

Answer 35

process of articulating more than one phoneme at a time

Question 36

what is segmentation?

Answer 36

the process of dividing the speech signal into component words

Question 37

example of segmentation?

Answer 37

when we spell the word dog, we separate it into its three separate sounds: /d/-/o/-/g/

Question 38

cohort model example?

Answer 38

discriminating between Crocodile and Dial, the point of recognition to discriminate between the two words comes at the /d/ in crocodile which is much earlier than the /l/ sound in Dial.